Vision transformers for dense prediction: A survey